1 /* Copyright (c) 2008, 2009 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
20 #include <arpa/inet.h>
24 #include <openflow/openflow.h>
29 #include <sys/socket.h>
30 #include <sys/types.h>
37 #include "dynamic-string.h"
41 #include "mac-learning.h"
44 #include "ofp-print.h"
47 #include "poll-loop.h"
48 #include "port-array.h"
49 #include "proc-net-compat.h"
51 #include "secchan/ofproto.h"
52 #include "socket-util.h"
59 #include "vconn-ssl.h"
60 #include "xenserver.h"
63 #define THIS_MODULE VLM_bridge
71 extern uint64_t mgmt_id;
74 struct port *port; /* Containing port. */
75 size_t port_ifidx; /* Index within containing port. */
77 char *name; /* Host network device name. */
78 int dp_ifidx; /* Index within kernel datapath. */
80 uint8_t mac[ETH_ADDR_LEN]; /* Ethernet address (all zeros if unknowns). */
82 tag_type tag; /* Tag associated with this interface. */
83 bool enabled; /* May be chosen for flows? */
84 long long delay_expires; /* Time after which 'enabled' may change. */
87 #define BOND_MASK 0xff
89 int iface_idx; /* Index of assigned iface, or -1 if none. */
90 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
91 tag_type iface_tag; /* Tag associated with iface_idx. */
94 #define MAX_MIRRORS 32
95 typedef uint32_t mirror_mask_t;
96 #define MIRROR_MASK_C(X) UINT32_C(X)
97 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
99 struct bridge *bridge;
103 /* Selection criteria. */
104 struct svec src_ports;
105 struct svec dst_ports;
110 struct port *out_port;
114 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
116 struct bridge *bridge;
118 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
119 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. */
122 /* An ordinary bridge port has 1 interface.
123 * A bridge port for bonding has at least 2 interfaces. */
124 struct iface **ifaces;
125 size_t n_ifaces, allocated_ifaces;
128 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
129 int active_iface; /* Ifidx on which bcasts accepted, or -1. */
130 tag_type active_iface_tag; /* Tag for bcast flows. */
131 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
132 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
134 /* Port mirroring info. */
135 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
136 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
137 bool is_mirror_output_port; /* Does port mirroring send frames here? */
139 /* Spanning tree info. */
140 enum stp_state stp_state; /* Always STP_FORWARDING if STP not in use. */
141 tag_type stp_state_tag; /* Tag for STP state change. */
144 #define DP_MAX_PORTS 255
146 struct list node; /* Node in global list of bridges. */
147 char *name; /* User-specified arbitrary name. */
148 struct mac_learning *ml; /* MAC learning table, or null not to learn. */
149 bool sent_config_request; /* Successfully sent config request? */
150 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
152 /* Support for remote controllers. */
153 char *controller; /* NULL if there is no remote controller;
154 * "discover" to do controller discovery;
155 * otherwise a vconn name. */
157 /* OpenFlow switch processing. */
158 struct ofproto *ofproto; /* OpenFlow switch. */
160 /* Kernel datapath information. */
161 struct dpif *dpif; /* Datapath. */
162 struct port_array ifaces; /* Indexed by kernel datapath port number. */
166 size_t n_ports, allocated_ports;
169 bool has_bonded_ports;
170 long long int bond_next_rebalance;
175 /* Flow statistics gathering. */
176 time_t next_stats_request;
178 /* Port mirroring. */
179 struct mirror *mirrors[MAX_MIRRORS];
183 long long int stp_last_tick;
186 /* List of all bridges. */
187 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
189 /* Maximum number of datapaths. */
190 enum { DP_MAX = 256 };
192 static struct bridge *bridge_create(const char *name);
193 static void bridge_destroy(struct bridge *);
194 static struct bridge *bridge_lookup(const char *name);
195 static int bridge_run_one(struct bridge *);
196 static void bridge_reconfigure_one(struct bridge *);
197 static void bridge_reconfigure_controller(struct bridge *);
198 static void bridge_get_all_ifaces(const struct bridge *, struct svec *ifaces);
199 static void bridge_fetch_dp_ifaces(struct bridge *);
200 static void bridge_flush(struct bridge *);
201 static void bridge_pick_local_hw_addr(struct bridge *,
202 uint8_t ea[ETH_ADDR_LEN],
203 const char **devname);
204 static uint64_t bridge_pick_datapath_id(struct bridge *,
205 const uint8_t bridge_ea[ETH_ADDR_LEN],
206 const char *devname);
207 static uint64_t dpid_from_hash(const void *, size_t nbytes);
209 static void bond_init(void);
210 static void bond_run(struct bridge *);
211 static void bond_wait(struct bridge *);
212 static void bond_rebalance_port(struct port *);
213 static void bond_send_learning_packets(struct port *);
215 static void port_create(struct bridge *, const char *name);
216 static void port_reconfigure(struct port *);
217 static void port_destroy(struct port *);
218 static struct port *port_lookup(const struct bridge *, const char *name);
219 static struct iface *port_lookup_iface(const struct port *, const char *name);
220 static struct port *port_from_dp_ifidx(const struct bridge *,
222 static void port_update_bond_compat(struct port *);
223 static void port_update_vlan_compat(struct port *);
225 static void mirror_create(struct bridge *, const char *name);
226 static void mirror_destroy(struct mirror *);
227 static void mirror_reconfigure(struct bridge *);
228 static void mirror_reconfigure_one(struct mirror *);
229 static bool vlan_is_mirrored(const struct mirror *, int vlan);
231 static void brstp_reconfigure(struct bridge *);
232 static void brstp_adjust_timers(struct bridge *);
233 static void brstp_run(struct bridge *);
234 static void brstp_wait(struct bridge *);
236 static void iface_create(struct port *, const char *name);
237 static void iface_destroy(struct iface *);
238 static struct iface *iface_lookup(const struct bridge *, const char *name);
239 static struct iface *iface_from_dp_ifidx(const struct bridge *,
242 /* Hooks into ofproto processing. */
243 static struct ofhooks bridge_ofhooks;
245 /* Public functions. */
247 /* Adds the name of each interface used by a bridge, including local and
248 * internal ports, to 'svec'. */
250 bridge_get_ifaces(struct svec *svec)
252 struct bridge *br, *next;
255 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
256 for (i = 0; i < br->n_ports; i++) {
257 struct port *port = br->ports[i];
259 for (j = 0; j < port->n_ifaces; j++) {
260 struct iface *iface = port->ifaces[j];
261 if (iface->dp_ifidx < 0) {
262 VLOG_ERR("%s interface not in datapath %s, ignoring",
263 iface->name, dpif_name(br->dpif));
265 if (iface->dp_ifidx != ODPP_LOCAL) {
266 svec_add(svec, iface->name);
274 /* The caller must already have called cfg_read(). */
283 for (i = 0; i < DP_MAX; i++) {
287 sprintf(devname, "dp%d", i);
288 retval = dpif_open(devname, &dpif);
290 char dpif_name[IF_NAMESIZE];
291 if (dpif_port_get_name(dpif, ODPP_LOCAL,
292 dpif_name, sizeof dpif_name)
293 || !cfg_has("bridge.%s.port", dpif_name)) {
297 } else if (retval != ENODEV) {
298 VLOG_ERR("failed to delete datapath dp%d: %s",
299 i, strerror(retval));
303 bridge_reconfigure();
308 config_string_change(const char *key, char **valuep)
310 const char *value = cfg_get_string(0, "%s", key);
311 if (value && (!*valuep || strcmp(value, *valuep))) {
313 *valuep = xstrdup(value);
321 bridge_configure_ssl(void)
323 /* XXX SSL should be configurable on a per-bridge basis.
324 * XXX should be possible to de-configure SSL. */
325 static char *private_key_file;
326 static char *certificate_file;
327 static char *cacert_file;
330 if (config_string_change("ssl.private-key", &private_key_file)) {
331 vconn_ssl_set_private_key_file(private_key_file);
334 if (config_string_change("ssl.certificate", &certificate_file)) {
335 vconn_ssl_set_certificate_file(certificate_file);
338 /* We assume that even if the filename hasn't changed, if the CA cert
339 * file has been removed, that we want to move back into
340 * boot-strapping mode. This opens a small security hole, because
341 * the old certificate will still be trusted until vSwitch is
342 * restarted. We may want to address this in vconn's SSL library. */
343 if (config_string_change("ssl.ca-cert", &cacert_file)
344 || (stat(cacert_file, &s) && errno == ENOENT)) {
345 vconn_ssl_set_ca_cert_file(cacert_file,
346 cfg_get_bool(0, "ssl.bootstrap-ca-cert"));
352 bridge_reconfigure(void)
354 struct svec old_br, new_br, raw_new_br;
355 struct bridge *br, *next;
358 COVERAGE_INC(bridge_reconfigure);
360 /* Collect old bridges. */
362 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
363 svec_add(&old_br, br->name);
366 /* Collect new bridges. */
367 svec_init(&raw_new_br);
368 cfg_get_subsections(&raw_new_br, "bridge");
370 for (i = 0; i < raw_new_br.n; i++) {
371 const char *name = raw_new_br.names[i];
372 if (!strncmp(name, "dp", 2) && isdigit(name[2])) {
373 VLOG_ERR("%s is not a valid bridge name (bridges may not be "
374 "named \"dp\" followed by a digit)", name);
376 svec_add(&new_br, name);
379 svec_destroy(&raw_new_br);
381 /* Get rid of deleted bridges and add new bridges. */
384 assert(svec_is_unique(&old_br));
385 assert(svec_is_unique(&new_br));
386 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
387 if (!svec_contains(&new_br, br->name)) {
391 for (i = 0; i < new_br.n; i++) {
392 const char *name = new_br.names[i];
393 if (!svec_contains(&old_br, name)) {
397 svec_destroy(&old_br);
398 svec_destroy(&new_br);
402 bridge_configure_ssl();
405 /* Reconfigure all bridges. */
406 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
407 bridge_reconfigure_one(br);
410 /* Add and delete ports on all datapaths.
412 * The kernel will reject any attempt to add a given port to a datapath if
413 * that port already belongs to a different datapath, so we must do all
414 * port deletions before any port additions. */
415 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
416 struct odp_port *dpif_ports;
418 struct svec want_ifaces;
420 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
421 bridge_get_all_ifaces(br, &want_ifaces);
422 for (i = 0; i < n_dpif_ports; i++) {
423 const struct odp_port *p = &dpif_ports[i];
424 if (!svec_contains(&want_ifaces, p->devname)
425 && strcmp(p->devname, br->name)) {
426 int retval = dpif_port_del(br->dpif, p->port);
428 VLOG_ERR("failed to remove %s interface from %s: %s",
429 p->devname, dpif_name(br->dpif),
434 svec_destroy(&want_ifaces);
437 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
438 struct odp_port *dpif_ports;
440 struct svec cur_ifaces, want_ifaces, add_ifaces;
442 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
443 svec_init(&cur_ifaces);
444 for (i = 0; i < n_dpif_ports; i++) {
445 svec_add(&cur_ifaces, dpif_ports[i].devname);
448 svec_sort_unique(&cur_ifaces);
449 bridge_get_all_ifaces(br, &want_ifaces);
450 svec_diff(&want_ifaces, &cur_ifaces, &add_ifaces, NULL, NULL);
452 for (i = 0; i < add_ifaces.n; i++) {
453 const char *if_name = add_ifaces.names[i];
454 int internal = cfg_get_bool(0, "iface.%s.internal", if_name);
455 int flags = internal ? ODP_PORT_INTERNAL : 0;
456 int error = dpif_port_add(br->dpif, if_name, flags, NULL);
457 if (error == EXFULL) {
458 VLOG_ERR("ran out of valid port numbers on %s",
459 dpif_name(br->dpif));
462 VLOG_ERR("failed to add %s interface to %s: %s",
463 if_name, dpif_name(br->dpif), strerror(error));
466 svec_destroy(&cur_ifaces);
467 svec_destroy(&want_ifaces);
468 svec_destroy(&add_ifaces);
470 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
473 struct iface *local_iface = NULL;
475 uint8_t engine_type, engine_id;
476 bool add_id_to_iface = false;
477 struct svec nf_hosts;
479 bridge_fetch_dp_ifaces(br);
480 for (i = 0; i < br->n_ports; ) {
481 struct port *port = br->ports[i];
483 for (j = 0; j < port->n_ifaces; ) {
484 struct iface *iface = port->ifaces[j];
485 if (iface->dp_ifidx < 0) {
486 VLOG_ERR("%s interface not in %s, dropping",
487 iface->name, dpif_name(br->dpif));
488 iface_destroy(iface);
490 if (iface->dp_ifidx == ODPP_LOCAL) {
493 VLOG_DBG("%s has interface %s on port %d",
495 iface->name, iface->dp_ifidx);
499 if (!port->n_ifaces) {
500 VLOG_ERR("%s port has no interfaces, dropping", port->name);
507 /* Pick local port hardware address, datapath ID. */
508 bridge_pick_local_hw_addr(br, ea, &devname);
510 int error = netdev_nodev_set_etheraddr(local_iface->name, ea);
512 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
513 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
514 "Ethernet address: %s",
515 br->name, strerror(error));
519 dpid = bridge_pick_datapath_id(br, ea, devname);
520 ofproto_set_datapath_id(br->ofproto, dpid);
522 /* Set NetFlow configuration on this bridge. */
523 dpif_get_netflow_ids(br->dpif, &engine_type, &engine_id);
524 if (cfg_has("netflow.%s.engine-type", br->name)) {
525 engine_type = cfg_get_int(0, "netflow.%s.engine-type",
528 if (cfg_has("netflow.%s.engine-id", br->name)) {
529 engine_id = cfg_get_int(0, "netflow.%s.engine-id", br->name);
531 if (cfg_has("netflow.%s.add-id-to-iface", br->name)) {
532 add_id_to_iface = cfg_get_bool(0, "netflow.%s.add-id-to-iface",
535 if (add_id_to_iface && engine_id > 0x7f) {
536 VLOG_WARN("bridge %s: netflow port mangling may conflict with "
537 "another vswitch, choose an engine id less than 128",
540 if (add_id_to_iface && br->n_ports > 0x1ff) {
541 VLOG_WARN("bridge %s: netflow port mangling will conflict with "
542 "another port when 512 or more ports are used",
545 svec_init(&nf_hosts);
546 cfg_get_all_keys(&nf_hosts, "netflow.%s.host", br->name);
547 if (ofproto_set_netflow(br->ofproto, &nf_hosts, engine_type,
548 engine_id, add_id_to_iface)) {
549 VLOG_ERR("bridge %s: problem setting netflow collectors",
553 /* Update the controller and related settings. It would be more
554 * straightforward to call this from bridge_reconfigure_one(), but we
555 * can't do it there for two reasons. First, and most importantly, at
556 * that point we don't know the dp_ifidx of any interfaces that have
557 * been added to the bridge (because we haven't actually added them to
558 * the datapath). Second, at that point we haven't set the datapath ID
559 * yet; when a controller is configured, resetting the datapath ID will
560 * immediately disconnect from the controller, so it's better to set
561 * the datapath ID before the controller. */
562 bridge_reconfigure_controller(br);
564 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
565 for (i = 0; i < br->n_ports; i++) {
566 struct port *port = br->ports[i];
567 port_update_vlan_compat(port);
570 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
571 brstp_reconfigure(br);
576 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
577 const char **devname)
579 uint64_t requested_ea;
585 /* Did the user request a particular MAC? */
586 requested_ea = cfg_get_mac(0, "bridge.%s.mac", br->name);
588 eth_addr_from_uint64(requested_ea, ea);
589 if (eth_addr_is_multicast(ea)) {
590 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
591 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
592 } else if (eth_addr_is_zero(ea)) {
593 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
599 /* Otherwise choose the minimum MAC address among all of the interfaces.
600 * (Xen uses FE:FF:FF:FF:FF:FF for virtual interfaces so this will get the
601 * MAC of the physical interface in such an environment.) */
602 memset(ea, 0xff, sizeof ea);
603 for (i = 0; i < br->n_ports; i++) {
604 struct port *port = br->ports[i];
605 if (port->is_mirror_output_port) {
608 for (j = 0; j < port->n_ifaces; j++) {
609 struct iface *iface = port->ifaces[j];
610 uint8_t iface_ea[ETH_ADDR_LEN];
611 if (iface->dp_ifidx == ODPP_LOCAL
612 || cfg_get_bool(0, "iface.%s.internal", iface->name)) {
615 error = netdev_nodev_get_etheraddr(iface->name, iface_ea);
617 if (!eth_addr_is_multicast(iface_ea) &&
618 !eth_addr_is_reserved(iface_ea) &&
619 !eth_addr_is_zero(iface_ea) &&
620 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0) {
621 memcpy(ea, iface_ea, ETH_ADDR_LEN);
622 *devname = iface->name;
625 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
626 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
627 iface->name, strerror(error));
631 if (eth_addr_is_multicast(ea) || eth_addr_is_vif(ea)) {
632 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
634 VLOG_WARN("bridge %s: using default bridge Ethernet "
635 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
637 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
638 br->name, ETH_ADDR_ARGS(ea));
642 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
643 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
644 * a network device, then that network device's name must be passed in as
645 * 'devname'; if 'bridge_ea' was derived some other way, then 'devname' must be
646 * passed in as a null pointer. */
648 bridge_pick_datapath_id(struct bridge *br,
649 const uint8_t bridge_ea[ETH_ADDR_LEN],
653 * The procedure for choosing a bridge MAC address will, in the most
654 * ordinary case, also choose a unique MAC that we can use as a datapath
655 * ID. In some special cases, though, multiple bridges will end up with
656 * the same MAC address. This is OK for the bridges, but it will confuse
657 * the OpenFlow controller, because each datapath needs a unique datapath
660 * Datapath IDs must be unique. It is also very desirable that they be
661 * stable from one run to the next, so that policy set on a datapath
666 dpid = cfg_get_dpid(0, "bridge.%s.datapath-id", br->name);
673 if (!netdev_get_vlan_vid(devname, &vlan)) {
675 * A bridge whose MAC address is taken from a VLAN network device
676 * (that is, a network device created with vconfig(8) or similar
677 * tool) will have the same MAC address as a bridge on the VLAN
678 * device's physical network device.
680 * Handle this case by hashing the physical network device MAC
681 * along with the VLAN identifier.
683 uint8_t buf[ETH_ADDR_LEN + 2];
684 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
685 buf[ETH_ADDR_LEN] = vlan >> 8;
686 buf[ETH_ADDR_LEN + 1] = vlan;
687 return dpid_from_hash(buf, sizeof buf);
690 * Assume that this bridge's MAC address is unique, since it
691 * doesn't fit any of the cases we handle specially.
696 * A purely internal bridge, that is, one that has no non-virtual
697 * network devices on it at all, is more difficult because it has no
698 * natural unique identifier at all.
700 * When the host is a XenServer, we handle this case by hashing the
701 * host's UUID with the name of the bridge. Names of bridges are
702 * persistent across XenServer reboots, although they can be reused if
703 * an internal network is destroyed and then a new one is later
704 * created, so this is fairly effective.
706 * When the host is not a XenServer, we punt by using a random MAC
707 * address on each run.
709 const char *host_uuid = xenserver_get_host_uuid();
711 char *combined = xasprintf("%s,%s", host_uuid, br->name);
712 dpid = dpid_from_hash(combined, strlen(combined));
718 return eth_addr_to_uint64(bridge_ea);
722 dpid_from_hash(const void *data, size_t n)
724 uint8_t hash[SHA1_DIGEST_SIZE];
726 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
727 sha1_bytes(data, n, hash);
728 eth_addr_mark_random(hash);
729 return eth_addr_to_uint64(hash);
735 struct bridge *br, *next;
739 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
740 int error = bridge_run_one(br);
742 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
743 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
744 "forcing reconfiguration", br->name);
758 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
759 ofproto_wait(br->ofproto);
760 if (br->controller) {
765 mac_learning_wait(br->ml);
772 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
773 * configuration changes. */
775 bridge_flush(struct bridge *br)
777 COVERAGE_INC(bridge_flush);
780 mac_learning_flush(br->ml);
784 /* Bridge reconfiguration functions. */
786 static struct bridge *
787 bridge_create(const char *name)
792 assert(!bridge_lookup(name));
793 br = xcalloc(1, sizeof *br);
795 error = dpif_create(name, &br->dpif);
796 if (error == EEXIST) {
797 error = dpif_open(name, &br->dpif);
799 VLOG_ERR("datapath %s already exists but cannot be opened: %s",
800 name, strerror(error));
804 dpif_flow_flush(br->dpif);
806 VLOG_ERR("failed to create datapath %s: %s", name, strerror(error));
811 error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto);
813 VLOG_ERR("failed to create switch %s: %s", name, strerror(error));
814 dpif_delete(br->dpif);
815 dpif_close(br->dpif);
820 br->name = xstrdup(name);
821 br->ml = mac_learning_create();
822 br->sent_config_request = false;
823 eth_addr_random(br->default_ea);
825 port_array_init(&br->ifaces);
828 br->bond_next_rebalance = time_msec() + 10000;
830 list_push_back(&all_bridges, &br->node);
832 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
838 bridge_destroy(struct bridge *br)
843 while (br->n_ports > 0) {
844 port_destroy(br->ports[br->n_ports - 1]);
846 list_remove(&br->node);
847 error = dpif_delete(br->dpif);
848 if (error && error != ENOENT) {
849 VLOG_ERR("failed to delete %s: %s",
850 dpif_name(br->dpif), strerror(error));
852 dpif_close(br->dpif);
853 ofproto_destroy(br->ofproto);
854 free(br->controller);
855 mac_learning_destroy(br->ml);
856 port_array_destroy(&br->ifaces);
863 static struct bridge *
864 bridge_lookup(const char *name)
868 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
869 if (!strcmp(br->name, name)) {
877 bridge_exists(const char *name)
879 return bridge_lookup(name) ? true : false;
883 bridge_get_datapathid(const char *name)
885 struct bridge *br = bridge_lookup(name);
886 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
890 bridge_run_one(struct bridge *br)
894 error = ofproto_run1(br->ofproto);
900 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
905 error = ofproto_run2(br->ofproto, br->flush);
912 bridge_get_controller(const struct bridge *br)
914 const char *controller;
916 controller = cfg_get_string(0, "bridge.%s.controller", br->name);
918 controller = cfg_get_string(0, "mgmt.controller");
920 return controller && controller[0] ? controller : NULL;
924 bridge_reconfigure_one(struct bridge *br)
926 struct svec old_ports, new_ports, ifaces;
927 struct svec listeners, old_listeners;
928 struct svec snoops, old_snoops;
931 /* Collect old ports. */
932 svec_init(&old_ports);
933 for (i = 0; i < br->n_ports; i++) {
934 svec_add(&old_ports, br->ports[i]->name);
936 svec_sort(&old_ports);
937 assert(svec_is_unique(&old_ports));
939 /* Collect new ports. */
940 svec_init(&new_ports);
941 cfg_get_all_keys(&new_ports, "bridge.%s.port", br->name);
942 svec_sort(&new_ports);
943 if (bridge_get_controller(br) && !svec_contains(&new_ports, br->name)) {
944 svec_add(&new_ports, br->name);
945 svec_sort(&new_ports);
947 if (!svec_is_unique(&new_ports)) {
948 VLOG_WARN("bridge %s: %s specified twice as bridge port",
949 br->name, svec_get_duplicate(&new_ports));
950 svec_unique(&new_ports);
953 ofproto_set_mgmt_id(br->ofproto, mgmt_id);
955 /* Get rid of deleted ports and add new ports. */
956 for (i = 0; i < br->n_ports; ) {
957 struct port *port = br->ports[i];
958 if (!svec_contains(&new_ports, port->name)) {
964 for (i = 0; i < new_ports.n; i++) {
965 const char *name = new_ports.names[i];
966 if (!svec_contains(&old_ports, name)) {
967 port_create(br, name);
970 svec_destroy(&old_ports);
971 svec_destroy(&new_ports);
973 /* Reconfigure all ports. */
974 for (i = 0; i < br->n_ports; i++) {
975 port_reconfigure(br->ports[i]);
978 /* Check and delete duplicate interfaces. */
980 for (i = 0; i < br->n_ports; ) {
981 struct port *port = br->ports[i];
982 for (j = 0; j < port->n_ifaces; ) {
983 struct iface *iface = port->ifaces[j];
984 if (svec_contains(&ifaces, iface->name)) {
985 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
987 br->name, iface->name, port->name);
988 iface_destroy(iface);
990 svec_add(&ifaces, iface->name);
995 if (!port->n_ifaces) {
996 VLOG_ERR("%s port has no interfaces, dropping", port->name);
1002 svec_destroy(&ifaces);
1004 /* Delete all flows if we're switching from connected to standalone or vice
1005 * versa. (XXX Should we delete all flows if we are switching from one
1006 * controller to another?) */
1008 /* Configure OpenFlow management listeners. */
1009 svec_init(&listeners);
1010 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1012 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1013 ovs_rundir, br->name));
1014 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1015 svec_clear(&listeners);
1017 svec_sort_unique(&listeners);
1019 svec_init(&old_listeners);
1020 ofproto_get_listeners(br->ofproto, &old_listeners);
1021 svec_sort_unique(&old_listeners);
1023 if (!svec_equal(&listeners, &old_listeners)) {
1024 ofproto_set_listeners(br->ofproto, &listeners);
1026 svec_destroy(&listeners);
1027 svec_destroy(&old_listeners);
1029 /* Configure OpenFlow controller connection snooping. */
1031 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1033 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1034 ovs_rundir, br->name));
1035 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1036 svec_clear(&snoops);
1038 svec_sort_unique(&snoops);
1040 svec_init(&old_snoops);
1041 ofproto_get_snoops(br->ofproto, &old_snoops);
1042 svec_sort_unique(&old_snoops);
1044 if (!svec_equal(&snoops, &old_snoops)) {
1045 ofproto_set_snoops(br->ofproto, &snoops);
1047 svec_destroy(&snoops);
1048 svec_destroy(&old_snoops);
1050 mirror_reconfigure(br);
1054 bridge_reconfigure_controller(struct bridge *br)
1056 char *pfx = xasprintf("bridge.%s.controller", br->name);
1057 const char *controller;
1059 controller = bridge_get_controller(br);
1060 if ((br->controller != NULL) != (controller != NULL)) {
1061 ofproto_flush_flows(br->ofproto);
1063 free(br->controller);
1064 br->controller = controller ? xstrdup(controller) : NULL;
1067 const char *fail_mode;
1068 int max_backoff, probe;
1069 int rate_limit, burst_limit;
1071 if (!strcmp(controller, "discover")) {
1072 ofproto_set_discovery(br->ofproto, true,
1073 cfg_get_string(0, "%s.accept-regex", pfx),
1074 cfg_get_bool(0, "%s.update-resolv.conf",
1077 struct netdev *netdev;
1081 in_band = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
1083 || cfg_get_bool(0, "%s.in-band", pfx));
1084 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1085 ofproto_set_in_band(br->ofproto, in_band);
1087 error = netdev_open(br->name, NETDEV_ETH_TYPE_NONE, &netdev);
1089 if (cfg_is_valid(CFG_IP | CFG_REQUIRED, "%s.ip", pfx)) {
1090 struct in_addr ip, mask, gateway;
1091 ip.s_addr = cfg_get_ip(0, "%s.ip", pfx);
1092 mask.s_addr = cfg_get_ip(0, "%s.netmask", pfx);
1093 gateway.s_addr = cfg_get_ip(0, "%s.gateway", pfx);
1095 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1097 mask.s_addr = guess_netmask(ip.s_addr);
1099 if (!netdev_set_in4(netdev, ip, mask)) {
1100 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1102 br->name, IP_ARGS(&ip.s_addr),
1103 IP_ARGS(&mask.s_addr));
1106 if (gateway.s_addr) {
1107 if (!netdev_add_router(gateway)) {
1108 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1109 br->name, IP_ARGS(&gateway.s_addr));
1113 netdev_close(netdev);
1117 fail_mode = cfg_get_string(0, "%s.fail-mode", pfx);
1119 fail_mode = cfg_get_string(0, "mgmt.fail-mode");
1121 ofproto_set_failure(br->ofproto,
1123 || !strcmp(fail_mode, "standalone")
1124 || !strcmp(fail_mode, "open")));
1126 probe = cfg_get_int(0, "%s.inactivity-probe", pfx);
1128 probe = cfg_get_int(0, "mgmt.inactivity-probe");
1133 ofproto_set_probe_interval(br->ofproto, probe);
1135 max_backoff = cfg_get_int(0, "%s.max-backoff", pfx);
1137 max_backoff = cfg_get_int(0, "mgmt.max-backoff");
1142 ofproto_set_max_backoff(br->ofproto, max_backoff);
1144 rate_limit = cfg_get_int(0, "%s.rate-limit", pfx);
1146 rate_limit = cfg_get_int(0, "mgmt.rate-limit");
1148 burst_limit = cfg_get_int(0, "%s.burst-limit", pfx);
1150 burst_limit = cfg_get_int(0, "mgmt.burst-limit");
1152 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1154 ofproto_set_stp(br->ofproto, cfg_get_bool(0, "%s.stp", pfx));
1156 if (cfg_has("%s.commands.acl", pfx)) {
1157 struct svec command_acls;
1160 svec_init(&command_acls);
1161 cfg_get_all_strings(&command_acls, "%s.commands.acl", pfx);
1162 command_acl = svec_join(&command_acls, ",", "");
1164 ofproto_set_remote_execution(br->ofproto, command_acl,
1165 cfg_get_string(0, "%s.commands.dir",
1168 svec_destroy(&command_acls);
1171 ofproto_set_remote_execution(br->ofproto, NULL, NULL);
1174 union ofp_action action;
1177 /* Set up a flow that matches every packet and directs them to
1178 * OFPP_NORMAL (which goes to us). */
1179 memset(&action, 0, sizeof action);
1180 action.type = htons(OFPAT_OUTPUT);
1181 action.output.len = htons(sizeof action);
1182 action.output.port = htons(OFPP_NORMAL);
1183 memset(&flow, 0, sizeof flow);
1184 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1187 ofproto_set_in_band(br->ofproto, false);
1188 ofproto_set_max_backoff(br->ofproto, 1);
1189 ofproto_set_probe_interval(br->ofproto, 5);
1190 ofproto_set_failure(br->ofproto, false);
1191 ofproto_set_stp(br->ofproto, false);
1195 ofproto_set_controller(br->ofproto, br->controller);
1199 bridge_get_all_ifaces(const struct bridge *br, struct svec *ifaces)
1204 for (i = 0; i < br->n_ports; i++) {
1205 struct port *port = br->ports[i];
1206 for (j = 0; j < port->n_ifaces; j++) {
1207 struct iface *iface = port->ifaces[j];
1208 svec_add(ifaces, iface->name);
1212 assert(svec_is_unique(ifaces));
1215 /* For robustness, in case the administrator moves around datapath ports behind
1216 * our back, we re-check all the datapath port numbers here.
1218 * This function will set the 'dp_ifidx' members of interfaces that have
1219 * disappeared to -1, so only call this function from a context where those
1220 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1221 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1222 * datapath, which doesn't support UINT16_MAX+1 ports. */
1224 bridge_fetch_dp_ifaces(struct bridge *br)
1226 struct odp_port *dpif_ports;
1227 size_t n_dpif_ports;
1230 /* Reset all interface numbers. */
1231 for (i = 0; i < br->n_ports; i++) {
1232 struct port *port = br->ports[i];
1233 for (j = 0; j < port->n_ifaces; j++) {
1234 struct iface *iface = port->ifaces[j];
1235 iface->dp_ifidx = -1;
1238 port_array_clear(&br->ifaces);
1240 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1241 for (i = 0; i < n_dpif_ports; i++) {
1242 struct odp_port *p = &dpif_ports[i];
1243 struct iface *iface = iface_lookup(br, p->devname);
1245 if (iface->dp_ifidx >= 0) {
1246 VLOG_WARN("%s reported interface %s twice",
1247 dpif_name(br->dpif), p->devname);
1248 } else if (iface_from_dp_ifidx(br, p->port)) {
1249 VLOG_WARN("%s reported interface %"PRIu16" twice",
1250 dpif_name(br->dpif), p->port);
1252 port_array_set(&br->ifaces, p->port, iface);
1253 iface->dp_ifidx = p->port;
1260 /* Bridge packet processing functions. */
1263 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1265 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1268 static struct bond_entry *
1269 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1271 return &port->bond_hash[bond_hash(mac)];
1275 bond_choose_iface(const struct port *port)
1278 for (i = 0; i < port->n_ifaces; i++) {
1279 if (port->ifaces[i]->enabled) {
1287 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1288 uint16_t *dp_ifidx, tag_type *tags)
1290 struct iface *iface;
1292 assert(port->n_ifaces);
1293 if (port->n_ifaces == 1) {
1294 iface = port->ifaces[0];
1296 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1297 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1298 || !port->ifaces[e->iface_idx]->enabled) {
1299 /* XXX select interface properly. The current interface selection
1300 * is only good for testing the rebalancing code. */
1301 e->iface_idx = bond_choose_iface(port);
1302 if (e->iface_idx < 0) {
1303 *tags |= port->no_ifaces_tag;
1306 e->iface_tag = tag_create_random();
1308 *tags |= e->iface_tag;
1309 iface = port->ifaces[e->iface_idx];
1311 *dp_ifidx = iface->dp_ifidx;
1312 *tags |= iface->tag; /* Currently only used for bonding. */
1317 bond_link_status_update(struct iface *iface, bool carrier)
1319 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1320 struct port *port = iface->port;
1322 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1323 /* Nothing to do. */
1326 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1327 iface->name, carrier ? "detected" : "dropped");
1328 if (carrier == iface->enabled) {
1329 iface->delay_expires = LLONG_MAX;
1330 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1331 iface->name, carrier ? "disabled" : "enabled");
1333 int delay = carrier ? port->updelay : port->downdelay;
1334 iface->delay_expires = time_msec() + delay;
1337 "interface %s: will be %s if it stays %s for %d ms",
1339 carrier ? "enabled" : "disabled",
1340 carrier ? "up" : "down",
1347 bond_choose_active_iface(struct port *port)
1349 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1351 port->active_iface = bond_choose_iface(port);
1352 port->active_iface_tag = tag_create_random();
1353 if (port->active_iface >= 0) {
1354 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1355 port->name, port->ifaces[port->active_iface]->name);
1357 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1363 bond_enable_slave(struct iface *iface, bool enable)
1365 struct port *port = iface->port;
1366 struct bridge *br = port->bridge;
1368 iface->delay_expires = LLONG_MAX;
1369 if (enable == iface->enabled) {
1373 iface->enabled = enable;
1374 if (!iface->enabled) {
1375 VLOG_WARN("interface %s: enabled", iface->name);
1376 ofproto_revalidate(br->ofproto, iface->tag);
1377 if (iface->port_ifidx == port->active_iface) {
1378 ofproto_revalidate(br->ofproto,
1379 port->active_iface_tag);
1380 bond_choose_active_iface(port);
1382 bond_send_learning_packets(port);
1384 VLOG_WARN("interface %s: disabled", iface->name);
1385 if (port->active_iface < 0) {
1386 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1387 bond_choose_active_iface(port);
1388 bond_send_learning_packets(port);
1390 iface->tag = tag_create_random();
1395 bond_run(struct bridge *br)
1399 for (i = 0; i < br->n_ports; i++) {
1400 struct port *port = br->ports[i];
1401 if (port->n_ifaces < 2) {
1404 for (j = 0; j < port->n_ifaces; j++) {
1405 struct iface *iface = port->ifaces[j];
1406 if (time_msec() >= iface->delay_expires) {
1407 bond_enable_slave(iface, !iface->enabled);
1414 bond_wait(struct bridge *br)
1418 for (i = 0; i < br->n_ports; i++) {
1419 struct port *port = br->ports[i];
1420 if (port->n_ifaces < 2) {
1423 for (j = 0; j < port->n_ifaces; j++) {
1424 struct iface *iface = port->ifaces[j];
1425 if (iface->delay_expires != LLONG_MAX) {
1426 poll_timer_wait(iface->delay_expires - time_msec());
1433 set_dst(struct dst *p, const flow_t *flow,
1434 const struct port *in_port, const struct port *out_port,
1439 * XXX This uses too many tags: any broadcast flow will get one tag per
1440 * destination port, and thus a broadcast on a switch of any size is likely
1441 * to have all tag bits set. We should figure out a way to be smarter.
1443 * This is OK when STP is disabled, because stp_state_tag is 0 then. */
1444 *tags |= out_port->stp_state_tag;
1445 if (!(out_port->stp_state & (STP_DISABLED | STP_FORWARDING))) {
1449 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1450 : in_port->vlan >= 0 ? in_port->vlan
1451 : ntohs(flow->dl_vlan));
1452 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1456 swap_dst(struct dst *p, struct dst *q)
1458 struct dst tmp = *p;
1463 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1464 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1465 * that we push to the datapath. We could in fact fully sort the array by
1466 * vlan, but in most cases there are at most two different vlan tags so that's
1467 * possibly overkill.) */
1469 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1471 struct dst *first = dsts;
1472 struct dst *last = dsts + n_dsts;
1474 while (first != last) {
1476 * - All dsts < first have vlan == 'vlan'.
1477 * - All dsts >= last have vlan != 'vlan'.
1478 * - first < last. */
1479 while (first->vlan == vlan) {
1480 if (++first == last) {
1485 /* Same invariants, plus one additional:
1486 * - first->vlan != vlan.
1488 while (last[-1].vlan != vlan) {
1489 if (--last == first) {
1494 /* Same invariants, plus one additional:
1495 * - last[-1].vlan == vlan.*/
1496 swap_dst(first++, --last);
1501 mirror_mask_ffs(mirror_mask_t mask)
1503 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1508 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1509 const struct dst *test)
1512 for (i = 0; i < n_dsts; i++) {
1513 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1521 port_trunks_vlan(const struct port *port, uint16_t vlan)
1523 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1527 port_includes_vlan(const struct port *port, uint16_t vlan)
1529 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1533 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1534 const struct port *in_port, const struct port *out_port,
1535 struct dst dsts[], tag_type *tags)
1537 mirror_mask_t mirrors = in_port->src_mirrors;
1538 struct dst *dst = dsts;
1541 *tags |= in_port->stp_state_tag;
1542 if (out_port == FLOOD_PORT) {
1543 /* XXX use ODP_FLOOD if no vlans or bonding. */
1544 /* XXX even better, define each VLAN as a datapath port group */
1545 for (i = 0; i < br->n_ports; i++) {
1546 struct port *port = br->ports[i];
1547 if (port != in_port && port_includes_vlan(port, vlan)
1548 && !port->is_mirror_output_port
1549 && set_dst(dst, flow, in_port, port, tags)) {
1550 mirrors |= port->dst_mirrors;
1554 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1555 mirrors |= out_port->dst_mirrors;
1560 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1561 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1563 if (set_dst(dst, flow, in_port, m->out_port, tags)
1564 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1568 for (i = 0; i < br->n_ports; i++) {
1569 struct port *port = br->ports[i];
1570 if (port_includes_vlan(port, m->out_vlan)
1571 && set_dst(dst, flow, in_port, port, tags)
1572 && !dst_is_duplicate(dsts, dst - dsts, dst))
1574 if (port->vlan < 0) {
1575 dst->vlan = m->out_vlan;
1577 if (dst->dp_ifidx == flow->in_port
1578 && dst->vlan == vlan) {
1579 /* Don't send out input port on same VLAN. */
1587 mirrors &= mirrors - 1;
1590 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
1595 print_dsts(const struct dst *dsts, size_t n)
1597 for (; n--; dsts++) {
1598 printf(">p%"PRIu16, dsts->dp_ifidx);
1599 if (dsts->vlan != OFP_VLAN_NONE) {
1600 printf("v%"PRIu16, dsts->vlan);
1606 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
1607 const struct port *in_port, const struct port *out_port,
1608 tag_type *tags, struct odp_actions *actions)
1610 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
1612 const struct dst *p;
1615 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags);
1617 cur_vlan = ntohs(flow->dl_vlan);
1618 for (p = dsts; p < &dsts[n_dsts]; p++) {
1619 union odp_action *a;
1620 if (p->vlan != cur_vlan) {
1621 if (p->vlan == OFP_VLAN_NONE) {
1622 odp_actions_add(actions, ODPAT_STRIP_VLAN);
1624 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
1625 a->vlan_vid.vlan_vid = htons(p->vlan);
1629 a = odp_actions_add(actions, ODPAT_OUTPUT);
1630 a->output.port = p->dp_ifidx;
1635 is_bcast_arp_reply(const flow_t *flow, const struct ofpbuf *packet)
1637 struct arp_eth_header *arp = (struct arp_eth_header *) packet->data;
1638 return (flow->dl_type == htons(ETH_TYPE_ARP)
1639 && eth_addr_is_broadcast(flow->dl_dst)
1640 && packet->size >= sizeof(struct arp_eth_header)
1641 && arp->ar_op == ARP_OP_REQUEST);
1644 /* If the composed actions may be applied to any packet in the given 'flow',
1645 * returns true. Otherwise, the actions should only be applied to 'packet', or
1646 * not at all, if 'packet' was NULL. */
1648 process_flow(struct bridge *br, const flow_t *flow,
1649 const struct ofpbuf *packet, struct odp_actions *actions,
1652 struct iface *in_iface;
1653 struct port *in_port;
1654 struct port *out_port = NULL; /* By default, drop the packet/flow. */
1657 /* Find the interface and port structure for the received packet. */
1658 in_iface = iface_from_dp_ifidx(br, flow->in_port);
1660 /* No interface? Something fishy... */
1661 if (packet != NULL) {
1662 /* Odd. A few possible reasons here:
1664 * - We deleted an interface but there are still a few packets
1665 * queued up from it.
1667 * - Someone externally added an interface (e.g. with "ovs-dpctl
1668 * add-if") that we don't know about.
1670 * - Packet arrived on the local port but the local port is not
1671 * one of our bridge ports.
1673 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1675 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1676 "interface %"PRIu16, br->name, flow->in_port);
1679 /* Return without adding any actions, to drop packets on this flow. */
1682 in_port = in_iface->port;
1684 /* Figure out what VLAN this packet belongs to.
1686 * Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
1687 * belongs to VLAN 0, so we should treat both cases identically. (In the
1688 * former case, the packet has an 802.1Q header that specifies VLAN 0,
1689 * presumably to allow a priority to be specified. In the latter case, the
1690 * packet does not have any 802.1Q header.) */
1691 vlan = ntohs(flow->dl_vlan);
1692 if (vlan == OFP_VLAN_NONE) {
1695 if (in_port->vlan >= 0) {
1697 /* XXX support double tagging? */
1698 if (packet != NULL) {
1699 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1700 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
1701 "packet received on port %s configured with "
1702 "implicit VLAN %"PRIu16,
1703 br->name, ntohs(flow->dl_vlan),
1704 in_port->name, in_port->vlan);
1708 vlan = in_port->vlan;
1710 if (!port_includes_vlan(in_port, vlan)) {
1711 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1712 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
1713 "packet received on port %s not configured for "
1715 br->name, vlan, in_port->name, vlan);
1720 /* Drop frames for ports that STP wants entirely killed (both for
1721 * forwarding and for learning). Later, after we do learning, we'll drop
1722 * the frames that STP wants to do learning but not forwarding on. */
1723 if (in_port->stp_state & (STP_LISTENING | STP_BLOCKING)) {
1727 /* Drop frames for reserved multicast addresses. */
1728 if (eth_addr_is_reserved(flow->dl_dst)) {
1732 /* Drop frames on ports reserved for mirroring. */
1733 if (in_port->is_mirror_output_port) {
1734 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1735 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
1736 "which is reserved exclusively for mirroring",
1737 br->name, in_port->name);
1741 /* Multicast (and broadcast) packets on bonds need special attention, to
1742 * avoid receiving duplicates. */
1743 if (in_port->n_ifaces > 1 && eth_addr_is_multicast(flow->dl_dst)) {
1744 *tags |= in_port->active_iface_tag;
1745 if (in_port->active_iface != in_iface->port_ifidx) {
1746 /* Drop all multicast packets on inactive slaves. */
1749 /* Drop all multicast packets for which we have learned a different
1750 * input port, because we probably sent the packet on one slaves
1751 * and got it back on the active slave. Broadcast ARP replies are
1752 * an exception to this rule: the host has moved to another
1754 int src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
1755 if (src_idx != -1 && src_idx != in_port->port_idx) {
1757 if (!is_bcast_arp_reply(flow, packet)) {
1761 /* No way to know whether it's an ARP reply, because the
1762 * flow entry doesn't include enough information and we
1763 * don't have a packet. Punt. */
1771 out_port = FLOOD_PORT;
1775 /* Learn source MAC (but don't try to learn from revalidation). */
1777 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
1778 vlan, in_port->port_idx);
1780 /* The log messages here could actually be useful in debugging,
1781 * so keep the rate limit relatively high. */
1782 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
1784 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1785 "on port %s in VLAN %d",
1786 br->name, ETH_ADDR_ARGS(flow->dl_src),
1787 in_port->name, vlan);
1788 ofproto_revalidate(br->ofproto, rev_tag);
1792 /* Determine output port. */
1793 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
1795 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
1796 out_port = br->ports[out_port_idx];
1800 /* Don't send packets out their input ports. Don't forward frames that STP
1801 * wants us to discard. */
1802 if (in_port == out_port || in_port->stp_state == STP_LEARNING) {
1807 compose_actions(br, flow, vlan, in_port, out_port, tags, actions);
1810 * We send out only a single packet, instead of setting up a flow, if the
1811 * packet is an ARP directed to broadcast that arrived on a bonded
1812 * interface. In such a situation ARP requests and replies must be handled
1813 * differently, but OpenFlow unfortunately can't distinguish them.
1815 return (in_port->n_ifaces < 2
1816 || flow->dl_type != htons(ETH_TYPE_ARP)
1817 || !eth_addr_is_broadcast(flow->dl_dst));
1820 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
1823 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
1824 const struct ofp_phy_port *opp,
1827 struct bridge *br = br_;
1828 struct iface *iface;
1831 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
1837 if (reason == OFPPR_DELETE) {
1838 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
1839 br->name, iface->name);
1840 iface_destroy(iface);
1841 if (!port->n_ifaces) {
1842 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1843 br->name, port->name);
1849 memcpy(iface->mac, opp->hw_addr, ETH_ADDR_LEN);
1850 if (port->n_ifaces > 1) {
1851 bool up = !(opp->state & OFPPS_LINK_DOWN);
1852 bond_link_status_update(iface, up);
1853 port_update_bond_compat(port);
1859 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
1860 struct odp_actions *actions, tag_type *tags, void *br_)
1862 struct bridge *br = br_;
1865 if (flow->dl_type == htons(OFP_DL_TYPE_NOT_ETH_TYPE)
1866 && eth_addr_equals(flow->dl_dst, stp_eth_addr)) {
1867 brstp_receive(br, flow, payload);
1872 COVERAGE_INC(bridge_process_flow);
1873 return process_flow(br, flow, packet, actions, tags);
1877 bridge_account_flow_ofhook_cb(const flow_t *flow,
1878 const union odp_action *actions,
1879 size_t n_actions, unsigned long long int n_bytes,
1882 struct bridge *br = br_;
1883 const union odp_action *a;
1885 if (!br->has_bonded_ports) {
1889 for (a = actions; a < &actions[n_actions]; a++) {
1890 if (a->type == ODPAT_OUTPUT) {
1891 struct port *port = port_from_dp_ifidx(br, a->output.port);
1892 if (port && port->n_ifaces >= 2) {
1893 struct bond_entry *e = lookup_bond_entry(port, flow->dl_src);
1894 e->tx_bytes += n_bytes;
1901 bridge_account_checkpoint_ofhook_cb(void *br_)
1903 struct bridge *br = br_;
1906 if (!br->has_bonded_ports) {
1910 /* The current ofproto implementation calls this callback at least once a
1911 * second, so this timer implementation is sufficient. */
1912 if (time_msec() < br->bond_next_rebalance) {
1915 br->bond_next_rebalance = time_msec() + 10000;
1917 for (i = 0; i < br->n_ports; i++) {
1918 struct port *port = br->ports[i];
1919 if (port->n_ifaces > 1) {
1920 bond_rebalance_port(port);
1925 static struct ofhooks bridge_ofhooks = {
1926 bridge_port_changed_ofhook_cb,
1927 bridge_normal_ofhook_cb,
1928 bridge_account_flow_ofhook_cb,
1929 bridge_account_checkpoint_ofhook_cb,
1932 /* Bonding functions. */
1934 /* Statistics for a single interface on a bonded port, used for load-based
1935 * bond rebalancing. */
1936 struct slave_balance {
1937 struct iface *iface; /* The interface. */
1938 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
1940 /* All the "bond_entry"s that are assigned to this interface, in order of
1941 * increasing tx_bytes. */
1942 struct bond_entry **hashes;
1946 /* Sorts pointers to pointers to bond_entries in ascending order by the
1947 * interface to which they are assigned, and within a single interface in
1948 * ascending order of bytes transmitted. */
1950 compare_bond_entries(const void *a_, const void *b_)
1952 const struct bond_entry *const *ap = a_;
1953 const struct bond_entry *const *bp = b_;
1954 const struct bond_entry *a = *ap;
1955 const struct bond_entry *b = *bp;
1956 if (a->iface_idx != b->iface_idx) {
1957 return a->iface_idx > b->iface_idx ? 1 : -1;
1958 } else if (a->tx_bytes != b->tx_bytes) {
1959 return a->tx_bytes > b->tx_bytes ? 1 : -1;
1965 /* Sorts slave_balances so that enabled ports come first, and otherwise in
1966 * *descending* order by number of bytes transmitted. */
1968 compare_slave_balance(const void *a_, const void *b_)
1970 const struct slave_balance *a = a_;
1971 const struct slave_balance *b = b_;
1972 if (a->iface->enabled != b->iface->enabled) {
1973 return a->iface->enabled ? -1 : 1;
1974 } else if (a->tx_bytes != b->tx_bytes) {
1975 return a->tx_bytes > b->tx_bytes ? -1 : 1;
1982 swap_bals(struct slave_balance *a, struct slave_balance *b)
1984 struct slave_balance tmp = *a;
1989 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
1990 * given that 'p' (and only 'p') might be in the wrong location.
1992 * This function invalidates 'p', since it might now be in a different memory
1995 resort_bals(struct slave_balance *p,
1996 struct slave_balance bals[], size_t n_bals)
1999 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2000 swap_bals(p, p - 1);
2002 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2003 swap_bals(p, p + 1);
2009 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2011 if (VLOG_IS_DBG_ENABLED()) {
2012 struct ds ds = DS_EMPTY_INITIALIZER;
2013 const struct slave_balance *b;
2015 for (b = bals; b < bals + n_bals; b++) {
2019 ds_put_char(&ds, ',');
2021 ds_put_format(&ds, " %s %"PRIu64"kB",
2022 b->iface->name, b->tx_bytes / 1024);
2024 if (!b->iface->enabled) {
2025 ds_put_cstr(&ds, " (disabled)");
2027 if (b->n_hashes > 0) {
2028 ds_put_cstr(&ds, " (");
2029 for (i = 0; i < b->n_hashes; i++) {
2030 const struct bond_entry *e = b->hashes[i];
2032 ds_put_cstr(&ds, " + ");
2034 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2035 e - port->bond_hash, e->tx_bytes / 1024);
2037 ds_put_cstr(&ds, ")");
2040 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2045 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2047 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2048 struct bond_entry *hash)
2050 struct port *port = from->iface->port;
2051 uint64_t delta = hash->tx_bytes;
2053 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2054 "from %s to %s (now carrying %"PRIu64"kB and "
2055 "%"PRIu64"kB load, respectively)",
2056 port->name, delta / 1024, hash - port->bond_hash,
2057 from->iface->name, to->iface->name,
2058 (from->tx_bytes - delta) / 1024,
2059 (to->tx_bytes + delta) / 1024);
2061 /* Delete element from from->hashes.
2063 * We don't bother to add the element to to->hashes because not only would
2064 * it require more work, the only purpose it would be to allow that hash to
2065 * be migrated to another slave in this rebalancing run, and there is no
2066 * point in doing that. */
2067 if (from->hashes[0] == hash) {
2070 int i = hash - from->hashes[0];
2071 memmove(from->hashes + i, from->hashes + i + 1,
2072 (from->n_hashes - (i + 1)) * sizeof *from->hashes);
2076 /* Shift load away from 'from' to 'to'. */
2077 from->tx_bytes -= delta;
2078 to->tx_bytes += delta;
2080 /* Arrange for flows to be revalidated. */
2081 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2082 hash->iface_idx = to->iface->port_ifidx;
2083 hash->iface_tag = tag_create_random();
2087 bond_rebalance_port(struct port *port)
2089 struct slave_balance bals[DP_MAX_PORTS];
2091 struct bond_entry *hashes[BOND_MASK + 1];
2092 struct slave_balance *b, *from, *to;
2093 struct bond_entry *e;
2096 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2097 * descending order of tx_bytes, so that bals[0] represents the most
2098 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2101 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2102 * array for each slave_balance structure, we sort our local array of
2103 * hashes in order by slave, so that all of the hashes for a given slave
2104 * become contiguous in memory, and then we point each 'hashes' members of
2105 * a slave_balance structure to the start of a contiguous group. */
2106 n_bals = port->n_ifaces;
2107 for (b = bals; b < &bals[n_bals]; b++) {
2108 b->iface = port->ifaces[b - bals];
2113 for (i = 0; i <= BOND_MASK; i++) {
2114 hashes[i] = &port->bond_hash[i];
2116 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2117 for (i = 0; i <= BOND_MASK; i++) {
2119 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2120 b = &bals[e->iface_idx];
2121 b->tx_bytes += e->tx_bytes;
2123 b->hashes = &hashes[i];
2128 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2129 log_bals(bals, n_bals, port);
2131 /* Discard slaves that aren't enabled (which were sorted to the back of the
2132 * array earlier). */
2133 while (!bals[n_bals - 1].iface->enabled) {
2140 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2141 to = &bals[n_bals - 1];
2142 for (from = bals; from < to; ) {
2143 uint64_t overload = from->tx_bytes - to->tx_bytes;
2144 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2145 /* The extra load on 'from' (and all less-loaded slaves), compared
2146 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2147 * it is less than ~1Mbps. No point in rebalancing. */
2149 } else if (from->n_hashes == 1) {
2150 /* 'from' only carries a single MAC hash, so we can't shift any
2151 * load away from it, even though we want to. */
2154 /* 'from' is carrying significantly more load than 'to', and that
2155 * load is split across at least two different hashes. Pick a hash
2156 * to migrate to 'to' (the least-loaded slave), given that doing so
2157 * must not cause 'to''s load to exceed 'from''s load.
2159 * The sort order we use means that we prefer to shift away the
2160 * smallest hashes instead of the biggest ones. There is little
2161 * reason behind this decision; we could use the opposite sort
2162 * order to shift away big hashes ahead of small ones. */
2165 for (i = 0; i < from->n_hashes; i++) {
2166 uint64_t delta = from->hashes[i]->tx_bytes;
2167 if (to->tx_bytes + delta < from->tx_bytes - delta) {
2171 if (i < from->n_hashes) {
2172 bond_shift_load(from, to, from->hashes[i]);
2174 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2175 * point to different slave_balance structures. It is only
2176 * valid to do these two operations in a row at all because we
2177 * know that 'from' will not move past 'to' and vice versa. */
2178 resort_bals(from, bals, n_bals);
2179 resort_bals(to, bals, n_bals);
2186 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2187 * historical data to decay to <1% in 7 rebalancing runs. */
2188 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2194 bond_send_learning_packets(struct port *port)
2196 struct bridge *br = port->bridge;
2197 struct mac_entry *e;
2198 struct ofpbuf packet;
2199 int error, n_packets, n_errors;
2201 if (!port->n_ifaces || port->active_iface < 0 || !br->ml) {
2205 ofpbuf_init(&packet, 128);
2206 error = n_packets = n_errors = 0;
2207 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2208 static const char s[] = "Open vSwitch Bond Failover";
2209 union ofp_action actions[2], *a;
2210 struct eth_header *eth;
2211 struct llc_snap_header *llc_snap;
2217 if (e->port == port->port_idx
2218 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2222 /* Compose packet to send. */
2223 ofpbuf_clear(&packet);
2224 eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
2225 llc_snap = ofpbuf_put_zeros(&packet, LLC_SNAP_HEADER_LEN);
2226 ofpbuf_put(&packet, s, sizeof s); /* Includes null byte. */
2227 ofpbuf_put(&packet, e->mac, ETH_ADDR_LEN);
2229 memcpy(eth->eth_dst, eth_addr_broadcast, ETH_ADDR_LEN);
2230 memcpy(eth->eth_src, e->mac, ETH_ADDR_LEN);
2231 eth->eth_type = htons(packet.size - ETH_HEADER_LEN);
2233 llc_snap->llc.llc_dsap = LLC_DSAP_SNAP;
2234 llc_snap->llc.llc_ssap = LLC_SSAP_SNAP;
2235 llc_snap->llc.llc_cntl = LLC_CNTL_SNAP;
2236 memcpy(llc_snap->snap.snap_org, "\x00\x23\x20", 3);
2237 llc_snap->snap.snap_type = htons(0xf177); /* Random number. */
2239 /* Compose actions. */
2240 memset(actions, 0, sizeof actions);
2243 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2244 a->vlan_vid.len = htons(sizeof *a);
2245 a->vlan_vid.vlan_vid = htons(e->vlan);
2248 a->output.type = htons(OFPAT_OUTPUT);
2249 a->output.len = htons(sizeof *a);
2250 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2255 flow_extract(&packet, ODPP_NONE, &flow);
2256 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2263 ofpbuf_uninit(&packet);
2266 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2267 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2268 "packets, last error was: %s",
2269 port->name, n_errors, n_packets, strerror(error));
2271 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2272 port->name, n_packets);
2276 /* Bonding unixctl user interface functions. */
2279 bond_unixctl_list(struct unixctl_conn *conn, const char *args UNUSED)
2281 struct ds ds = DS_EMPTY_INITIALIZER;
2282 const struct bridge *br;
2284 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2286 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2289 for (i = 0; i < br->n_ports; i++) {
2290 const struct port *port = br->ports[i];
2291 if (port->n_ifaces > 1) {
2294 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2295 for (j = 0; j < port->n_ifaces; j++) {
2296 const struct iface *iface = port->ifaces[j];
2298 ds_put_cstr(&ds, ", ");
2300 ds_put_cstr(&ds, iface->name);
2302 ds_put_char(&ds, '\n');
2306 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2310 static struct port *
2311 bond_find(const char *name)
2313 const struct bridge *br;
2315 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2318 for (i = 0; i < br->n_ports; i++) {
2319 struct port *port = br->ports[i];
2320 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2329 bond_unixctl_show(struct unixctl_conn *conn, const char *args)
2331 struct ds ds = DS_EMPTY_INITIALIZER;
2332 const struct port *port;
2335 port = bond_find(args);
2337 unixctl_command_reply(conn, 501, "no such bond");
2341 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2342 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2343 ds_put_format(&ds, "next rebalance: %lld ms\n",
2344 port->bridge->bond_next_rebalance - time_msec());
2345 for (j = 0; j < port->n_ifaces; j++) {
2346 const struct iface *iface = port->ifaces[j];
2347 struct bond_entry *be;
2350 ds_put_format(&ds, "slave %s: %s\n",
2351 iface->name, iface->enabled ? "enabled" : "disabled");
2352 if (j == port->active_iface) {
2353 ds_put_cstr(&ds, "\tactive slave\n");
2355 if (iface->delay_expires != LLONG_MAX) {
2356 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2357 iface->enabled ? "downdelay" : "updelay",
2358 iface->delay_expires - time_msec());
2362 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2363 int hash = be - port->bond_hash;
2364 struct mac_entry *me;
2366 if (be->iface_idx != j) {
2370 ds_put_format(&ds, "\thash %d: %lld kB load\n",
2371 hash, be->tx_bytes / 1024);
2374 if (!port->bridge->ml) {
2378 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2379 &port->bridge->ml->lrus) {
2382 if (bond_hash(me->mac) == hash
2383 && me->port != port->port_idx
2384 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2385 && dp_ifidx == iface->dp_ifidx)
2387 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2388 ETH_ADDR_ARGS(me->mac));
2393 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2398 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_)
2400 char *args = (char *) args_;
2401 char *save_ptr = NULL;
2402 char *bond_s, *hash_s, *slave_s;
2403 uint8_t mac[ETH_ADDR_LEN];
2405 struct iface *iface;
2406 struct bond_entry *entry;
2409 bond_s = strtok_r(args, " ", &save_ptr);
2410 hash_s = strtok_r(NULL, " ", &save_ptr);
2411 slave_s = strtok_r(NULL, " ", &save_ptr);
2413 unixctl_command_reply(conn, 501,
2414 "usage: bond/migrate BOND HASH SLAVE");
2418 port = bond_find(bond_s);
2420 unixctl_command_reply(conn, 501, "no such bond");
2424 if (sscanf(hash_s, "%"SCNx8":%"SCNx8":%"SCNx8":%"SCNx8":%"SCNx8":%"SCNx8,
2425 &mac[0], &mac[1], &mac[2], &mac[3], &mac[4], &mac[5]) == 6) {
2426 hash = bond_hash(mac);
2427 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2428 hash = atoi(hash_s) & BOND_MASK;
2430 unixctl_command_reply(conn, 501, "bad hash");
2434 iface = port_lookup_iface(port, slave_s);
2436 unixctl_command_reply(conn, 501, "no such slave");
2440 if (!iface->enabled) {
2441 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2445 entry = &port->bond_hash[hash];
2446 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2447 entry->iface_idx = iface->port_ifidx;
2448 entry->iface_tag = tag_create_random();
2449 unixctl_command_reply(conn, 200, "migrated");
2453 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_)
2455 char *args = (char *) args_;
2456 char *save_ptr = NULL;
2457 char *bond_s, *slave_s;
2459 struct iface *iface;
2461 bond_s = strtok_r(args, " ", &save_ptr);
2462 slave_s = strtok_r(NULL, " ", &save_ptr);
2464 unixctl_command_reply(conn, 501,
2465 "usage: bond/set-active-slave BOND SLAVE");
2469 port = bond_find(bond_s);
2471 unixctl_command_reply(conn, 501, "no such bond");
2475 iface = port_lookup_iface(port, slave_s);
2477 unixctl_command_reply(conn, 501, "no such slave");
2481 if (!iface->enabled) {
2482 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2486 if (port->active_iface != iface->port_ifidx) {
2487 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2488 port->active_iface = iface->port_ifidx;
2489 port->active_iface_tag = tag_create_random();
2490 VLOG_INFO("port %s: active interface is now %s",
2491 port->name, iface->name);
2492 bond_send_learning_packets(port);
2493 unixctl_command_reply(conn, 200, "done");
2495 unixctl_command_reply(conn, 200, "no change");
2500 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2502 char *args = (char *) args_;
2503 char *save_ptr = NULL;
2504 char *bond_s, *slave_s;
2506 struct iface *iface;
2508 bond_s = strtok_r(args, " ", &save_ptr);
2509 slave_s = strtok_r(NULL, " ", &save_ptr);
2511 unixctl_command_reply(conn, 501,
2512 "usage: bond/enable/disable-slave BOND SLAVE");
2516 port = bond_find(bond_s);
2518 unixctl_command_reply(conn, 501, "no such bond");
2522 iface = port_lookup_iface(port, slave_s);
2524 unixctl_command_reply(conn, 501, "no such slave");
2528 bond_enable_slave(iface, enable);
2529 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
2533 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args)
2535 enable_slave(conn, args, true);
2539 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args)
2541 enable_slave(conn, args, false);
2547 unixctl_command_register("bond/list", bond_unixctl_list);
2548 unixctl_command_register("bond/show", bond_unixctl_show);
2549 unixctl_command_register("bond/migrate", bond_unixctl_migrate);
2550 unixctl_command_register("bond/set-active-slave",
2551 bond_unixctl_set_active_slave);
2552 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave);
2553 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave);
2556 /* Port functions. */
2559 port_create(struct bridge *br, const char *name)
2563 port = xcalloc(1, sizeof *port);
2565 port->port_idx = br->n_ports;
2567 port->trunks = NULL;
2568 port->name = xstrdup(name);
2569 port->active_iface = -1;
2570 port->stp_state = STP_DISABLED;
2571 port->stp_state_tag = 0;
2573 if (br->n_ports >= br->allocated_ports) {
2574 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
2577 br->ports[br->n_ports++] = port;
2579 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
2584 port_reconfigure(struct port *port)
2586 bool bonded = cfg_has_section("bonding.%s", port->name);
2587 struct svec old_ifaces, new_ifaces;
2588 unsigned long *trunks;
2592 /* Collect old and new interfaces. */
2593 svec_init(&old_ifaces);
2594 svec_init(&new_ifaces);
2595 for (i = 0; i < port->n_ifaces; i++) {
2596 svec_add(&old_ifaces, port->ifaces[i]->name);
2598 svec_sort(&old_ifaces);
2600 cfg_get_all_keys(&new_ifaces, "bonding.%s.slave", port->name);
2601 if (!new_ifaces.n) {
2602 VLOG_ERR("port %s: no interfaces specified for bonded port",
2604 } else if (new_ifaces.n == 1) {
2605 VLOG_WARN("port %s: only 1 interface specified for bonded port",
2609 port->updelay = cfg_get_int(0, "bonding.%s.updelay", port->name);
2610 if (port->updelay < 0) {
2613 port->downdelay = cfg_get_int(0, "bonding.%s.downdelay", port->name);
2614 if (port->downdelay < 0) {
2615 port->downdelay = 0;
2618 svec_init(&new_ifaces);
2619 svec_add(&new_ifaces, port->name);
2622 /* Get rid of deleted interfaces and add new interfaces. */
2623 for (i = 0; i < port->n_ifaces; i++) {
2624 struct iface *iface = port->ifaces[i];
2625 if (!svec_contains(&new_ifaces, iface->name)) {
2626 iface_destroy(iface);
2631 for (i = 0; i < new_ifaces.n; i++) {
2632 const char *name = new_ifaces.names[i];
2633 if (!svec_contains(&old_ifaces, name)) {
2634 iface_create(port, name);
2640 if (cfg_has("vlan.%s.tag", port->name)) {
2642 vlan = cfg_get_vlan(0, "vlan.%s.tag", port->name);
2643 if (vlan >= 0 && vlan <= 4095) {
2644 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
2647 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
2648 * they even work as-is. But they have not been tested. */
2649 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
2653 if (port->vlan != vlan) {
2655 bridge_flush(port->bridge);
2658 /* Get trunked VLANs. */
2661 size_t n_trunks, n_errors;
2664 trunks = bitmap_allocate(4096);
2665 n_trunks = cfg_count("vlan.%s.trunks", port->name);
2667 for (i = 0; i < n_trunks; i++) {
2668 int trunk = cfg_get_vlan(i, "vlan.%s.trunks", port->name);
2670 bitmap_set1(trunks, trunk);
2676 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
2677 port->name, n_trunks);
2679 if (n_errors == n_trunks) {
2681 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
2684 bitmap_set_multiple(trunks, 0, 4096, 1);
2687 if (cfg_has("vlan.%s.trunks", port->name)) {
2688 VLOG_ERR("ignoring vlan.%s.trunks in favor of vlan.%s.vlan",
2689 port->name, port->name);
2693 ? port->trunks != NULL
2694 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
2695 bridge_flush(port->bridge);
2697 bitmap_free(port->trunks);
2698 port->trunks = trunks;
2700 svec_destroy(&old_ifaces);
2701 svec_destroy(&new_ifaces);
2705 port_destroy(struct port *port)
2708 struct bridge *br = port->bridge;
2712 proc_net_compat_update_vlan(port->name, NULL, 0);
2714 for (i = 0; i < MAX_MIRRORS; i++) {
2715 struct mirror *m = br->mirrors[i];
2716 if (m && m->out_port == port) {
2721 while (port->n_ifaces > 0) {
2722 iface_destroy(port->ifaces[port->n_ifaces - 1]);
2725 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
2726 del->port_idx = port->port_idx;
2729 bitmap_free(port->trunks);
2736 static struct port *
2737 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
2739 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
2740 return iface ? iface->port : NULL;
2743 static struct port *
2744 port_lookup(const struct bridge *br, const char *name)
2748 for (i = 0; i < br->n_ports; i++) {
2749 struct port *port = br->ports[i];
2750 if (!strcmp(port->name, name)) {
2757 static struct iface *
2758 port_lookup_iface(const struct port *port, const char *name)
2762 for (j = 0; j < port->n_ifaces; j++) {
2763 struct iface *iface = port->ifaces[j];
2764 if (!strcmp(iface->name, name)) {
2772 port_update_bonding(struct port *port)
2774 if (port->n_ifaces < 2) {
2775 /* Not a bonded port. */
2776 if (port->bond_hash) {
2777 free(port->bond_hash);
2778 port->bond_hash = NULL;
2779 proc_net_compat_update_bond(port->name, NULL);
2782 if (!port->bond_hash) {
2785 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
2786 for (i = 0; i <= BOND_MASK; i++) {
2787 struct bond_entry *e = &port->bond_hash[i];
2791 port->no_ifaces_tag = tag_create_random();
2792 bond_choose_active_iface(port);
2794 port_update_bond_compat(port);
2799 port_update_bond_compat(struct port *port)
2801 struct compat_bond bond;
2804 if (port->n_ifaces < 2) {
2809 bond.updelay = port->updelay;
2810 bond.downdelay = port->downdelay;
2811 bond.n_slaves = port->n_ifaces;
2812 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
2813 for (i = 0; i < port->n_ifaces; i++) {
2814 struct iface *iface = port->ifaces[i];
2815 struct compat_bond_slave *slave = &bond.slaves[i];
2816 slave->name = iface->name;
2817 slave->up = ((iface->enabled && iface->delay_expires == LLONG_MAX) ||
2818 (!iface->enabled && iface->delay_expires != LLONG_MAX));
2822 memcpy(slave->mac, iface->mac, ETH_ADDR_LEN);
2824 proc_net_compat_update_bond(port->name, &bond);
2829 port_update_vlan_compat(struct port *port)
2831 struct bridge *br = port->bridge;
2832 char *vlandev_name = NULL;
2834 if (port->vlan > 0) {
2835 /* Figure out the name that the VLAN device should actually have, if it
2836 * existed. This takes some work because the VLAN device would not
2837 * have port->name in its name; rather, it would have the trunk port's
2838 * name, and 'port' would be attached to a bridge that also had the
2839 * VLAN device one of its ports. So we need to find a trunk port that
2840 * includes port->vlan.
2842 * There might be more than one candidate. This doesn't happen on
2843 * XenServer, so if it happens we just pick the first choice in
2844 * alphabetical order instead of creating multiple VLAN devices. */
2846 for (i = 0; i < br->n_ports; i++) {
2847 struct port *p = br->ports[i];
2848 if (port_trunks_vlan(p, port->vlan)
2850 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
2852 const uint8_t *ea = p->ifaces[0]->mac;
2853 if (!eth_addr_is_multicast(ea) &&
2854 !eth_addr_is_reserved(ea) &&
2855 !eth_addr_is_zero(ea)) {
2856 vlandev_name = p->name;
2861 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
2864 /* Interface functions. */
2867 iface_create(struct port *port, const char *name)
2869 struct iface *iface;
2871 iface = xcalloc(1, sizeof *iface);
2873 iface->port_ifidx = port->n_ifaces;
2874 iface->name = xstrdup(name);
2875 iface->dp_ifidx = -1;
2876 iface->tag = tag_create_random();
2877 iface->delay_expires = LLONG_MAX;
2879 netdev_nodev_get_etheraddr(name, iface->mac);
2880 netdev_nodev_get_carrier(name, &iface->enabled);
2882 if (port->n_ifaces >= port->allocated_ifaces) {
2883 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
2884 sizeof *port->ifaces);
2886 port->ifaces[port->n_ifaces++] = iface;
2887 if (port->n_ifaces > 1) {
2888 port->bridge->has_bonded_ports = true;
2891 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
2893 port_update_bonding(port);
2894 bridge_flush(port->bridge);
2898 iface_destroy(struct iface *iface)
2901 struct port *port = iface->port;
2902 struct bridge *br = port->bridge;
2903 bool del_active = port->active_iface == iface->port_ifidx;
2906 if (iface->dp_ifidx >= 0) {
2907 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
2910 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
2911 del->port_ifidx = iface->port_ifidx;
2917 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2918 bond_choose_active_iface(port);
2919 bond_send_learning_packets(port);
2922 port_update_bonding(port);
2923 bridge_flush(port->bridge);
2927 static struct iface *
2928 iface_lookup(const struct bridge *br, const char *name)
2932 for (i = 0; i < br->n_ports; i++) {
2933 struct port *port = br->ports[i];
2934 for (j = 0; j < port->n_ifaces; j++) {
2935 struct iface *iface = port->ifaces[j];
2936 if (!strcmp(iface->name, name)) {
2944 static struct iface *
2945 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
2947 return port_array_get(&br->ifaces, dp_ifidx);
2950 /* Port mirroring. */
2953 mirror_reconfigure(struct bridge *br)
2955 struct svec old_mirrors, new_mirrors;
2958 /* Collect old and new mirrors. */
2959 svec_init(&old_mirrors);
2960 svec_init(&new_mirrors);
2961 cfg_get_subsections(&new_mirrors, "mirror.%s", br->name);
2962 for (i = 0; i < MAX_MIRRORS; i++) {
2963 if (br->mirrors[i]) {
2964 svec_add(&old_mirrors, br->mirrors[i]->name);
2968 /* Get rid of deleted mirrors and add new mirrors. */
2969 svec_sort(&old_mirrors);
2970 assert(svec_is_unique(&old_mirrors));
2971 svec_sort(&new_mirrors);
2972 assert(svec_is_unique(&new_mirrors));
2973 for (i = 0; i < MAX_MIRRORS; i++) {
2974 struct mirror *m = br->mirrors[i];
2975 if (m && !svec_contains(&new_mirrors, m->name)) {
2979 for (i = 0; i < new_mirrors.n; i++) {
2980 const char *name = new_mirrors.names[i];
2981 if (!svec_contains(&old_mirrors, name)) {
2982 mirror_create(br, name);
2985 svec_destroy(&old_mirrors);
2986 svec_destroy(&new_mirrors);
2988 /* Reconfigure all mirrors. */
2989 for (i = 0; i < MAX_MIRRORS; i++) {
2990 if (br->mirrors[i]) {
2991 mirror_reconfigure_one(br->mirrors[i]);
2995 /* Update port reserved status. */
2996 for (i = 0; i < br->n_ports; i++) {
2997 br->ports[i]->is_mirror_output_port = false;
2999 for (i = 0; i < MAX_MIRRORS; i++) {
3000 struct mirror *m = br->mirrors[i];
3001 if (m && m->out_port) {
3002 m->out_port->is_mirror_output_port = true;
3008 mirror_create(struct bridge *br, const char *name)
3013 for (i = 0; ; i++) {
3014 if (i >= MAX_MIRRORS) {
3015 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3016 "cannot create %s", br->name, MAX_MIRRORS, name);
3019 if (!br->mirrors[i]) {
3024 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3027 br->mirrors[i] = m = xcalloc(1, sizeof *m);
3030 m->name = xstrdup(name);
3031 svec_init(&m->src_ports);
3032 svec_init(&m->dst_ports);
3040 mirror_destroy(struct mirror *m)
3043 struct bridge *br = m->bridge;
3046 for (i = 0; i < br->n_ports; i++) {
3047 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3048 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3051 svec_destroy(&m->src_ports);
3052 svec_destroy(&m->dst_ports);
3055 m->bridge->mirrors[m->idx] = NULL;
3063 prune_ports(struct mirror *m, struct svec *ports)
3068 svec_sort_unique(ports);
3071 for (i = 0; i < ports->n; i++) {
3072 const char *name = ports->names[i];
3073 if (port_lookup(m->bridge, name)) {
3074 svec_add(&tmp, name);
3076 VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s",
3077 m->bridge->name, m->name, name);
3080 svec_swap(ports, &tmp);
3085 prune_vlans(struct mirror *m, struct svec *vlan_strings, int **vlans)
3089 /* This isn't perfect: it won't combine "0" and "00", and the textual sort
3090 * order won't give us numeric sort order. But that's good enough for what
3091 * we need right now. */
3092 svec_sort_unique(vlan_strings);
3094 *vlans = xmalloc(sizeof *vlans * vlan_strings->n);
3096 for (i = 0; i < vlan_strings->n; i++) {
3097 const char *name = vlan_strings->names[i];
3099 if (!str_to_int(name, 10, &vlan) || vlan < 0 || vlan > 4095) {
3100 VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s",
3101 m->bridge->name, m->name, name);
3103 (*vlans)[n_vlans++] = vlan;
3110 vlan_is_mirrored(const struct mirror *m, int vlan)
3114 for (i = 0; i < m->n_vlans; i++) {
3115 if (m->vlans[i] == vlan) {
3123 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3127 for (i = 0; i < m->n_vlans; i++) {
3128 if (port_trunks_vlan(p, m->vlans[i])) {
3136 mirror_reconfigure_one(struct mirror *m)
3138 char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name);
3139 struct svec src_ports, dst_ports, ports;
3140 struct svec vlan_strings;
3141 mirror_mask_t mirror_bit;
3142 const char *out_port_name;
3143 struct port *out_port;
3148 bool mirror_all_ports;
3150 /* Get output port. */
3151 out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port",
3152 m->bridge->name, m->name);
3153 if (out_port_name) {
3154 out_port = port_lookup(m->bridge, out_port_name);
3156 VLOG_ERR("%s.output.port: bridge %s does not have a port "
3157 "named %s", pfx, m->bridge->name, out_port_name);
3164 if (cfg_has("%s.output.vlan", pfx)) {
3165 VLOG_ERR("%s.output.port and %s.output.vlan both specified; "
3166 "ignoring %s.output.vlan", pfx, pfx, pfx);
3168 } else if (cfg_has("%s.output.vlan", pfx)) {
3170 out_vlan = cfg_get_vlan(0, "%s.output.vlan", pfx);
3172 VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, "
3173 "but exactly one is required; disabling port mirror %s",
3174 pfx, pfx, pfx, pfx);
3180 /* Get all the ports, and drop duplicates and ports that don't exist. */
3181 svec_init(&src_ports);
3182 svec_init(&dst_ports);
3184 cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx);
3185 cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx);
3186 cfg_get_all_keys(&ports, "%s.select.port", pfx);
3187 svec_append(&src_ports, &ports);
3188 svec_append(&dst_ports, &ports);
3189 svec_destroy(&ports);
3190 prune_ports(m, &src_ports);
3191 prune_ports(m, &dst_ports);
3193 /* Get all the vlans, and drop duplicate and invalid vlans. */
3194 svec_init(&vlan_strings);
3195 cfg_get_all_keys(&vlan_strings, "%s.select.vlan", pfx);
3196 n_vlans = prune_vlans(m, &vlan_strings, &vlans);
3197 svec_destroy(&vlan_strings);
3199 /* Update mirror data. */
3200 if (!svec_equal(&m->src_ports, &src_ports)
3201 || !svec_equal(&m->dst_ports, &dst_ports)
3202 || m->n_vlans != n_vlans
3203 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3204 || m->out_port != out_port
3205 || m->out_vlan != out_vlan) {
3206 bridge_flush(m->bridge);
3208 svec_swap(&m->src_ports, &src_ports);
3209 svec_swap(&m->dst_ports, &dst_ports);
3212 m->n_vlans = n_vlans;
3213 m->out_port = out_port;
3214 m->out_vlan = out_vlan;
3216 /* If no selection criteria have been given, mirror for all ports. */
3217 mirror_all_ports = (!m->src_ports.n) && (!m->dst_ports.n) && (!m->n_vlans);
3220 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3221 for (i = 0; i < m->bridge->n_ports; i++) {
3222 struct port *port = m->bridge->ports[i];
3224 if (mirror_all_ports
3225 || svec_contains(&m->src_ports, port->name)
3228 ? port_trunks_any_mirrored_vlan(m, port)
3229 : vlan_is_mirrored(m, port->vlan)))) {
3230 port->src_mirrors |= mirror_bit;
3232 port->src_mirrors &= ~mirror_bit;
3235 if (mirror_all_ports || svec_contains(&m->dst_ports, port->name)) {
3236 port->dst_mirrors |= mirror_bit;
3238 port->dst_mirrors &= ~mirror_bit;
3243 svec_destroy(&src_ports);
3244 svec_destroy(&dst_ports);
3248 /* Spanning tree protocol. */
3250 static void brstp_update_port_state(struct port *);
3253 brstp_send_bpdu(struct ofpbuf *pkt, int port_no, void *br_)
3255 struct bridge *br = br_;
3256 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3257 struct iface *iface = iface_from_dp_ifidx(br, port_no);
3259 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
3261 } else if (eth_addr_is_zero(iface->mac)) {
3262 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d with unknown MAC",
3265 union ofp_action action;
3266 struct eth_header *eth = pkt->l2;
3269 memcpy(eth->eth_src, iface->mac, ETH_ADDR_LEN);
3271 memset(&action, 0, sizeof action);
3272 action.type = htons(OFPAT_OUTPUT);
3273 action.output.len = htons(sizeof action);
3274 action.output.port = htons(port_no);
3276 flow_extract(pkt, ODPP_NONE, &flow);
3277 ofproto_send_packet(br->ofproto, &flow, &action, 1, pkt);
3283 brstp_reconfigure(struct bridge *br)
3287 if (!cfg_get_bool(0, "stp.%s.enabled", br->name)) {
3289 stp_destroy(br->stp);
3295 uint64_t bridge_address, bridge_id;
3296 int bridge_priority;
3298 bridge_address = cfg_get_mac(0, "stp.%s.address", br->name);
3299 if (!bridge_address) {
3301 bridge_address = (stp_get_bridge_id(br->stp)
3302 & ((UINT64_C(1) << 48) - 1));
3304 uint8_t mac[ETH_ADDR_LEN];
3305 eth_addr_random(mac);
3306 bridge_address = eth_addr_to_uint64(mac);
3310 if (cfg_is_valid(CFG_INT | CFG_REQUIRED, "stp.%s.priority",
3312 bridge_priority = cfg_get_int(0, "stp.%s.priority", br->name);
3314 bridge_priority = STP_DEFAULT_BRIDGE_PRIORITY;
3317 bridge_id = bridge_address | ((uint64_t) bridge_priority << 48);
3319 br->stp = stp_create(br->name, bridge_id, brstp_send_bpdu, br);
3320 br->stp_last_tick = time_msec();
3323 if (bridge_id != stp_get_bridge_id(br->stp)) {
3324 stp_set_bridge_id(br->stp, bridge_id);
3329 for (i = 0; i < br->n_ports; i++) {
3330 struct port *p = br->ports[i];
3332 struct stp_port *sp;
3333 int path_cost, priority;
3339 dp_ifidx = p->ifaces[0]->dp_ifidx;
3340 if (dp_ifidx < 0 || dp_ifidx >= STP_MAX_PORTS) {
3344 sp = stp_get_port(br->stp, dp_ifidx);
3345 enable = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
3346 "stp.%s.port.%s.enabled",
3348 || cfg_get_bool(0, "stp.%s.port.%s.enabled",
3349 br->name, p->name));
3350 if (p->is_mirror_output_port) {
3353 if (enable != (stp_port_get_state(sp) != STP_DISABLED)) {
3354 bridge_flush(br); /* Might not be necessary. */
3356 stp_port_enable(sp);
3358 stp_port_disable(sp);
3362 path_cost = cfg_get_int(0, "stp.%s.port.%s.path-cost",
3364 stp_port_set_path_cost(sp, path_cost ? path_cost : 19 /* XXX */);
3366 priority = (cfg_is_valid(CFG_INT | CFG_REQUIRED,
3367 "stp.%s.port.%s.priority",
3369 ? cfg_get_int(0, "stp.%s.port.%s.priority",
3371 : STP_DEFAULT_PORT_PRIORITY);
3372 stp_port_set_priority(sp, priority);
3375 brstp_adjust_timers(br);
3377 for (i = 0; i < br->n_ports; i++) {
3378 brstp_update_port_state(br->ports[i]);
3383 brstp_update_port_state(struct port *p)
3385 struct bridge *br = p->bridge;
3386 enum stp_state state;
3388 /* Figure out new state. */
3389 state = STP_DISABLED;
3390 if (br->stp && p->n_ifaces > 0) {
3391 int dp_ifidx = p->ifaces[0]->dp_ifidx;
3392 if (dp_ifidx >= 0 && dp_ifidx < STP_MAX_PORTS) {
3393 state = stp_port_get_state(stp_get_port(br->stp, dp_ifidx));
3398 if (p->stp_state != state) {
3399 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3400 VLOG_INFO_RL(&rl, "port %s: STP state changed from %s to %s",
3401 p->name, stp_state_name(p->stp_state),
3402 stp_state_name(state));
3403 if (p->stp_state == STP_DISABLED) {
3406 ofproto_revalidate(p->bridge->ofproto, p->stp_state_tag);
3408 p->stp_state = state;
3409 p->stp_state_tag = (p->stp_state == STP_DISABLED ? 0
3410 : tag_create_random());
3415 brstp_adjust_timers(struct bridge *br)
3417 int hello_time = cfg_get_int(0, "stp.%s.hello-time", br->name);
3418 int max_age = cfg_get_int(0, "stp.%s.max-age", br->name);
3419 int forward_delay = cfg_get_int(0, "stp.%s.forward-delay", br->name);
3421 stp_set_hello_time(br->stp, hello_time ? hello_time : 2000);
3422 stp_set_max_age(br->stp, max_age ? max_age : 20000);
3423 stp_set_forward_delay(br->stp, forward_delay ? forward_delay : 15000);
3427 brstp_run(struct bridge *br)
3430 long long int now = time_msec();
3431 long long int elapsed = now - br->stp_last_tick;
3432 struct stp_port *sp;
3435 stp_tick(br->stp, MIN(INT_MAX, elapsed));
3436 br->stp_last_tick = now;
3438 while (stp_get_changed_port(br->stp, &sp)) {
3439 struct port *p = port_from_dp_ifidx(br, stp_port_no(sp));
3441 brstp_update_port_state(p);
3448 brstp_wait(struct bridge *br)
3451 poll_timer_wait(1000);